The TBI gene expression microarray dataset was downloaded from the National Center for Biotechnology Information (NCBI) Gene Expression Omnibus (GEO, www.ncbi.nlm.nih.gov/geo/) database with the accession number GSE89866. This dataset, which was based on the GPL16791 Illumina HiSeq 2500 platform (Homo sapiens), contained TBI samples and no-TBI controls. Gene expression data were derived from the participants' serum. Blood samples were taken daily for three consecutive days after injury, and the sequencing results were compared with each participant's own baseline. A diagram of the analytical workflow was described in Figure 1.

NetworkAnalyst (https://www.networkanalyst.ca/) is an R-based analysis tool available online. We utilized NetworkAnalyst to compare gene expression profiles between different groups in GSE89866. Genes with an adjusted p-value < 0.05 were considered differentially expressed (DEGs). The volcano plot and heat map were created by an online tool (www.bioinformatics.com.cn).

Meanwhile, pyroptosis-related genes were extracted from GeneCards (Version 5.18.0, https://www.Genecards.org/). The search strings for pyroptosis in the GeneCard were listed in Supplementary Appendix 1. The differentially expressed pyroptosis-related genes (DEPGs) were obtained by overlapping DEGs and the pyroptosis-related genes. A Venn diagram was used to visualize the DEPGs (http://bioinformatics.psb.ugent.be/webtools/Venn/).

To explore the biological significance of the DEPGs, Gene Ontology (GO) annotation and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment analyses were performed by DAVID tools (https://david.ncifcrf.gov/). The entries with p-values < 0.05 were considered significant.

The proteins encoded by the DEPGs were used to construct the protein–protein interaction (PPI) network by Search Tool for the Retrieval of Interacting Genes (STRING, http://string-db.org). Cytoscape 3.6.1 (https://cytoscape.org) was utilized to visualize the results. Moreover, we adopted CytoHubba, a Cytoscape plug-in, to screen hub genes. Based on the results of four widely used algorithms—Maximum Clique Centrality (MCC), Edge Percolated Component (EPC), BottleNeck, and EcCentricity—the overlapped genes were identified as hub genes (3, 4). Visualization of the hub gene network was generated via GeneMAINA (http://genemania.org).

This study was conducted in accordance with the Helsinki Declaration (2013 revision) and was reviewed and approved by the Ethics Committee of Harrison International Peace Hospital. To protect patient privacy, all personal information was removed, and the review committee has waived the requirement of informed consent. The reporting of this study conforms to STROBE guidelines.

Patients diagnosed with sTBI (GCS score ≤ 8) were included in this study with the following inclusion criteria: ① a clear history of head trauma; ② age ≧ 18 years old; ③ blunt trauma; ④ hospital admission within 24 h after trauma; ⑤ Injury Severity Score (ISS) < 9 for injuries to other parts except the brain; ⑥ the CT image data after injury were complete. The exclusion criteria were as follows: ① previous neurological diseases such as stroke, Parkinson's disease, intracranial tumors, etc.; ② other specific diseases, such as severe infections within the last month; autoimmune diseases; hemorrhagic diseases or bleeding tendency; severe liver, kidney or heart disease; and malignant tumors; ③ pregnancy; ④ severe intellectual disability, mental illness, or other diseases that affect the patient's mental state.

The healthy controls, matched for sex and age, had no chronic diseases such as hypertension, diabetes, coronary heart disease, and no history of acute infection or medication in the month prior to testing. Routine blood tests, liver, and kidney function tests showed no obvious abnormalities.

Data related to patient baseline demographic and clinical characteristics extracted from medical records, including age, sex, smoking history, alcohol consumption history, past medical history, cause of injury, pupil parameters, and body mass index (BMI), etc. Pupil evaluation was referred to the previous literature (Supplementary Table 1) (6, 12, 13). Pupil reactivity was coded into three subgroups: (1) brisk (both pupils constricted identically with light shone into either eye alone); (2) sluggish (delayed pupil constriction in reaction to light); (3) fixed (no response to a light stimulus, regardless of which eye was being stimulated). Similarly, pupil size was coded into three subgroups: (1) normal (bilateral equal non-dilated pupil size); (2) anisocoric (unequal pupil sizes); (3) bilateral dilated.

At the same time, the Helsinki CT score was calculated from head CT scan images to assess radiological severity of head trauma (7). Positive radiological findings included epidural hematoma, subdural hematoma, intracerebral hematoma, intraventricular hemorrhage, lesion volume, and morphology of suprasellar cisterna. The scores range from −3 to 14, with higher scores indicating greater severity. All CT measurements were based on the Picture Archive Communication System (PACS). The specific scores were determined by the researcher who was not familiar with clinical data using thin slice CT scans of the head within 24 h after injury.

The candidate gene NFE2L2 was screened by aforementioned bioinformatics methods. To explore the relationship between NFE2L2 and cytokines, the expression levels of nuclear factor red blood cell 2 related factor 2 (Nrf2), IL1β and IL18 were detected by enzyme-linked immunosorbent assay (ELISA). Blood samples were obtained from the median cubital vein (both patients and healthy controls). All samples for Nrf2 testing were obtained at 24 h after injury. After centrifugation at 3,000 rpm for 10 min, serum samples were immediately subjected to the following analysis, or immediately stored at −80°C for further analysis.

According to the instructions of the ELISA kit (Elabscience; Wuhan, China), the same technician without knowledge of clinical and radiological information measured the serum levels of Nrf2, IL-1 β, and IL-18. In short, samples and standards were incubated in a pre-coated 96-well plate for 90 min, followed by a biotinylated antibody working solution for 1 h. After incubation, all holes were washed and incubated with antibodies conjugated with horseradish peroxidase for 30 min, followed by a substrate solution in the dark for 30 min. Finally, the termination solution was added to stop the reaction and the plate reading was immediately taken at 450 nm on the microplate reader (Thermo Fisher Scientific, USA).

The significance of differences in Nrf2 protein levels between sTBI and healthy controls was examined using the t-test. One-way analysis of variance or Kruskal–Wallis test was used for comparisons as appropriate. Spearman correlation test was used to investigate the correlation between serum Nrf2 protein and cytokines (IL-1β, IL-18). Analysis was performed using IBM SPSS Software version 25.0 (IBM Corp., Armonk, NY, USA) and GraphPad Prism 7 (GraphPad Software, San Diego, CA, USA). Results at p < 0.05 were considered statistically significant.

Based on the aforementioned screening criteria, 1,795 DEGs were obtained in GSE89866. The volcano plot and heat map are displayed in Figures 2A, B. Meanwhile, 98 pyroptosis-related genes were obtained from the GeneCards database (relevance score ≥2, Supplementary Table 2). The intersection of the above two gene sets was selected after Venn diagram analysis, and 11 overlapped genes were defined as DEPGs. The Venn diagram is shown in Figure 2C, and the detailed information on DEPGs is listed within Supplementary Table 3.

GO terms include biological processes (BP), cellular components (CC), and molecular functions (MF). As shown in Figure 2D, the top four significant GO terms were chosen according to p-value. In terms of GO analysis, DEPGs were mainly enriched in pyroptosis (p = 0.00000058), inflammatory response (p = 0.0000013) and sequence-specific DNA binding (p = 0.0086).

Likewise, as indicated by the results presented in Figure 2E, the significant enriched KEGG terms contained NOD-like receptor signaling pathway (p = 0.0000071) and Chemical carcinogenesis—reactive oxygen species (p = 0.013). These results demonstrated that chemokines and cytokines may play essential roles in the pathology of TBI. All of the GO terms and KEGG pathways were included in Supplementary Table 4.

The PPI network of DEPGs was analyzed according to the STRING database, including eight nodes and 13 edges (minimum required interaction score: 0.4, Figure 2F). To screen important nodes in the network, all nodes were ranked by the aforementioned methods in CytoHubba. Notably, the top four genes were consistent across methods, namely NFE2L2, TXNIP, NLRP3, and NLRC4 (Supplementary Table 5). The detailed description of the top four genes was provided in Table 1. To further explore these genes, we performed the functional interaction networks using the GeneMANIA database. Functional analysis revealed that these genes were involved in regulation of interleukin-1 beta production, inflammasome complex and regulation of response to cytokine stimulus (Figure 3). The results of functional analysis were consistent with previous studies (14, 15). Based on gene background screening and Nrf2 expression values recorded in medical records, we identified NFE2L2 as the candidate gene for further clinical study.

Table 2 summarized the demographic and clinical data of 81 patients with sTBI and 40 normal controls (NC) included in this study. The sTBI group included 46 males (56.8%) and 35 females (43.2%) with an age range of 18–89 years (mean ± standard difference, 52.48 ± 16.68 years), while the NC group consisted of 21 males (52.5%) and 19 females (47.5%) with an age range of 26–76 years (mean ± standard deviation, 52.18 ± 13.10 years). There were no statistically significant differences in age, sex, BMI, tobacco and alcohol exposure, and underlying diseases between sTBI patients and NC group.

Serum Nrf2 protein levels were measured using ELISA in 81 patients and 40 NCs. As shown in Figure 4A, Nrf2 protein levels were significantly higher in sTBI patients compared to NCs (p < 0.001).

To investigate the relationship between serum Nrf2 protein levels and disease severity of sTBI, the correlations between serum Nrf2 levels and GCS scores, pupil parameters, Helsinki CT scores were explored. The box plot showed that the levels of Nrf2 were reduced when the GCS scores was increased. There was negative relationship between the level of Nrf2 and the level of GCS (Figure 4B). In contrast, the levels of Nrf2 were increased when the levels of pupil parameters and Helsinki CT Scores were increased (Figures 4C, D). Spearman's rank correlation analysis showed that the levels of IL1β were increased when the level of Nrf2 was increased (Figure 4E). There was positive relationships between the level of Nrf2 and the level of IL1β. Similarly, the levels of IL18 were increased when the level of Nrf2 was increased (Figure 4F).